Research On Dynamic Load Balancing Strategy For Switch Migration

In software-defined networking,load balancing at the control layer has been one of the hot issues for research.Due to the dynamic changes of network traffic,uneven load distribution among controllers can occur,which affects network performance.Switch migration strategies can dynamically change the mapping relationship between controllers and switches,thus balancing the load gap between controllers and improving the load balancing performance of the control layer.The existing switch migration strategy is rigid in the selection of migration timing,resulting in too many migrations,which affects the stability of the network;secondly,the impact factors considered in the selection of migration targets are relatively single and have a limited impact on network performance;finally,the complexity of the policy is high and not suitable for large-scale networks.In response to these problems,two switch migration strategies are proposed in this paper,with the following main research content and contributions:(1)To address the problem of too many migrations,this paper proposes a migration target selection strategy based on switch load prediction.By introducing a load prediction component,the current and next load conditions are fully considered when selecting the switches to be migrated and the target controllers,and unnecessary migration actions are effectively avoided.The experimental results show that the load prediction component can effectively reduce the number of migrations.(2)To address the problem of relatively single consideration,this paper proposes a switch migration strategy based on two-sided matching.The strategy matches the switch to be migrated and the target controller in a satisfaction calculation by balancing several network performance metrics,and designs a one-to-one matching rule to avoid migration conflicts caused by parallel processing of overloaded controllers.The experimental results show that the strategy can maintain the load balancing performance at the control layer and reduce the response time of controllers while reducing the number of migrations and migration costs.(3)To address the problem that existing policies are highly complex and not applicable in large-scale networks,this paper proposes a switch migration strategy based on a multi-agent deep deterministic policy gradient algorithm.The strategy models the switch migration problem as a Markov game process of multi-agent,constructs a cooperative relationship between multi-agent,and designs the state space,action space and reward function according to the network definition and optimization objectives to achieve intelligent decision-making at the control layer.The experimental results show that the strategy has greater performance advantages in large-scale networks.To conclude,this paper proposes two switch migration strategies to address the problems of existing switch migration strategies.A network simulation environment based on Mininet and Ryu software has been built,and a large number of comparison experiments have been conducted under various real network topologies to verify the effectiveness and advantages of the proposed strategies.